Medical image processing apparatus, method and program

1. A medical image processing apparatus comprising: a continuous region extraction unit that receives a three-dimensional medical image representing bronchi, and that extracts a continuous region having a voxel value corresponding to an air region in the bronchi; a graph structure construction unit that constructs a graph structure corresponding to the continuous region; a linear structure extraction unit that extracts a plurality of linear structures representing fragments of the bronchi by receiving the three-dimensional medical image, and by analyzing a local density structure in a neighborhood of each point in the three-dimensional medical image; and a graph structure reconstruction unit that reconstructs a graph structure representing the bronchi by connecting a node constituting the graph structure corresponding to the continuous region and a node representing the linear structures to each other, and nodes representing the linear structures to each other by using a predetermined cost function, wherein different cost functions are defined for a first segment connecting the node constituting the graph structure corresponding to the continuous region and the node representing the linear structures and a second segment connecting the nodes representing the linear structures to each other, and wherein the cost function in the first segment is defined in such a manner that the first segment is more likely to be connected as a change in voxel values is smaller among first segments.

2. A medical image processing apparatus, as defined in claim 1 , wherein the linear structure extraction unit extracts the linear structures, and calculates the principal axis directions of the linear structures, and wherein the cost function in the second segment is defined in such a manner that the second segment is more likely to be connected as a change in the principal axis directions at nodes located at both ends of the second segment is smaller.

3. A medical image processing apparatus, as defined in claim 1 , the apparatus further comprising: a multi-resolution image generation unit that generates a plurality of images having different resolutions from each other from the three-dimensional medical image, wherein the continuous region extraction unit receives a lower resolution image of the generated plurality of images, and extracts the continuous region from the lower resolution image, and wherein the linear structure extraction unit receives a higher resolution image of the generated plurality of images, and performs extraction of the linear structure and calculation of the principal axis direction on the higher resolution image.

4. A medical image processing apparatus, as defined in claim 1 , wherein the continuous region extraction unit extracts the continuous region by using one of a graph-cut method, a region growing method, and a level set method.

5. A medical image processing apparatus, as defined in claim 1 , wherein the linear structure extraction unit performs eigenvalue analysis of Hessian matrix with respect to each point in the three-dimensional medical image.

6. A medical image processing apparatus, as defined in claim 1 , wherein the graph structure construction unit and the graph structure reconstruction unit construct graph structures by using a spanning tree algorithm.

7. A medical image processing method comprising the steps of: receiving a three-dimensional medical image representing bronchi, and extracting a continuous region having a voxel value corresponding to an air region in the bronchi; constructing a graph structure corresponding to the continuous region; extracting a plurality of linear structures representing fragments of the bronchi by receiving the three-dimensional medical image, and by analyzing a local density structure in a neighborhood of each point in the three-dimensional medical image; and reconstructing a graph structure representing the bronchi by connecting a node constituting the graph structure corresponding to the continuous region and a node representing the linear structures to each other, and nodes representing the linear structures to each other by using a predetermined cost function, wherein different cost functions are defined for a first segment connecting the node constituting the graph structure corresponding to the continuous region and the node representing the linear structures and a second segment connecting the nodes representing the linear structures to each other, and wherein the cost function in the first segment is defined in such a manner that the first segment is more likely to be connected as a change in voxel values is smaller among first segments.

8. A non-transitory computer-readable recording medium storing therein a medical image processing program for causing a computer to execute the steps of: receiving a three-dimensional medical image representing bronchi, and extracting a continuous region having a voxel value corresponding to an air region in the bronchi; constructing a graph structure corresponding to the continuous region; extracting a plurality of linear structures representing fragments of the bronchi by receiving the three-dimensional medical image, and by analyzing a local density structure in a neighborhood of each point in the three-dimensional medical image; and reconstructing a graph structure representing the bronchi by connecting a node constituting the graph structure corresponding to the continuous region and a node representing the linear structures to each other, and nodes representing the linear structures to each other by using a predetermined cost function, wherein different cost functions are defined for a first segment connecting the node constituting the graph structure corresponding to the continuous region and the node representing the linear structures and a second segment connecting the nodes representing the linear structures to each other, and wherein the cost function in the first segment is defined in such a manner that the first segment is more likely to be connected as a change in voxel values is smaller among first segments.